unclear: weak math-topic words
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 4 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron responds only weakly, mostly to technical words in mathematics text such as 'singular' and 'inversion'. Its pattern is faint and mixed, so it is hard to describe simply.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
The study of the real points of an algebraic variety is the subject of real algebraic geometry.⏎- A large part of singularity theory is devoted to the singularities of algebraic varieties.⏎- When computers became more common, a field named
-
plane belongs to an algebraic curve if its coordinates satisfy a given polynomial equation. Basic questions involve the study of the points of special interest like the singular points, the inflection points and the points at infinity. More advanced questions involve
-
not insignificant. The author compares these odds to other unlikely events and includes ideas on how the government could help lower these odds.⏎%0 Electronic Source %A Plait, Phil %D December 6, 2010 %T Bad
-
ID 5 volumes cannot be striped, concatenated, or mirrored.⏎Do not create a RAID 5 volume from a component that contains an existing file system. Doing so will erase the data during the RAID 5 initialization process.⏎A RAID 5
-
and even special effects in movies. It is essentially the evolution of a merger between a graphical user interface and Non-Uniform Rational B-Splines (NURBS) geometry, and currently functions for all major platforms.⏎Become
-
ordinary and the other is an antiparticle. With respect to symmetries and invariance principles, three special conservation laws have been described, associated with inversion or reversal of space, time, and charge.⏎Conservation laws are considered to
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately: the number of people who are in a group. The number of people who are in a group is 200. Now, the number of people in a group is 100. The number of people in a group is
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
a branch of mathematics studying zeros of polynomial equations. Modern algebraic geometry is based on more abstract techniques of abstract algebra, especially commutative algebra, with the language and the problems of geometry.⏎The main objects of
-
CC BY-NC-SA⏎Given an integer , the exponential sum version of the Gauss sum modulo is defined by⏎and the quadratic residue version is defined by⏎where is the Jacobi symbol, which has values
-
representability of via the Yoneda lemma. That is, any element induces, in natural bijective fashion, a natural transformation⏎and is universal if is an isomorphism.⏎Thus, universal elements are part and parcel of any discussion involving
-
both sides. Part A: Short Answer (1 problems, 12 points). 1. Let S and T be linear spaces and A : S T be a linear map. Say V and W are particular solutions of the equations A V
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L3:25 suppresses
- L3:461 excites
- L3:924 suppresses
- L3:52 suppresses
- L3:105 excites
- L0:2536 weakly excites
- L2:84 weakly excites
- L3:185 weakly suppresses
- L3:2272 weakly suppresses
- L3:1258 weakly excites
- L2:454 weakly excites
- L3:314 weakly excites
- L2:122 weakly excites
- L3:408 weakly excites
- L0:282 weakly suppresses
- L3:278 weakly excites
- L1:3029 weakly excites
- L1:2387 weakly excites
- L0:1955 weakly suppresses
- L3:703 weakly suppresses
- L1:307 weakly suppresses
- L3:421 weakly suppresses
- L0:1495 weakly suppresses
- L0:1475 weakly excites
Feeds later layers
- L8:312 weakly suppresses
- L8:383 weakly excites
- L5:2180 weakly excites
- L9:2326 weakly excites
- L8:3028 weakly suppresses
- L12:1870 weakly excites
- L9:932 weakly suppresses
- L10:659 weakly suppresses
- L16:1656 weakly excites
- L15:2913 weakly excites
- L20:1031 weakly excites
- L5:1128 weakly suppresses
- L9:878 weakly excites
- L12:2315 weakly excites
- L6:1371 weakly excites
- L15:873 weakly suppresses
- L11:2975 weakly excites
- L12:1341 weakly suppresses
- L18:2125 weakly suppresses
- L18:2437 weakly excites
- L22:1365 weakly excites
- L23:1645 weakly excites
- L7:1989 weakly excites
- L15:2561 weakly suppresses
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- TRACT
- 形式
- 大海
- SEG
- 树枝
- iona
- NONE
- 版本
- 的变化
- 效果
Suppresses
- her
- iland
- ivement
- .hibernate
- iven
- ernen
- initialised
- tipped
- PropertyInfo
- ement